Een Nederlandstalige versie van deze pagina is hier beschikbaar.

On October 28, Joos Buijs will defend his Ph.D. thesis entitled “Flexible Evolutionary Algorithms for Mining Structured Process Models”.

Preceding the defense there are several other meetings planned. Everyone is more than welcome to attend but registration beforehand is appreciated.

Day Programme

The global day programme on Tuesday October 28 2014 is as follows:

10:00 – 12:30 Scientific Workshop (Scientific Presentations in English) (MF 11/12)

13:30 – 15:30 CoSeLoG Meeting (Presentations in Dutch from CoSeLoG participants) (Zwarte Doos/Black Box room 2.03)

16:00 – 17:30 Defense Joos (Auditorium 4)

17:30 – 18:30 Reception

Locations

All events take place at Eindhoven University of Technology.

The scientific workshop takes place in the MetaForum building lecture room 11/12, which are on floor 4.

The location for the CoSeLoG meeting will be announced soon.

The defense of Joos will be in Auditorium room 4. The reception location will be announced at the defense.

More information about the campus of Eindhoven University of Technology can be found at http://www.tue.nl/en/university/about-the-university/accessibility-tue-campus/

Scientific Workshop

In the scientific workshop several members of the promotion comittee present their work. Although this is mainly targetted at researchers, everyone is invited to attend. The presentations will be given in English.

The agenda for the workshop is as follows:

10:00 - 10:05 opening

10:05 - 10:35 Metric learning and model interpretability (Barbara Hammer)

10:35 - 11:05 Business Process Deviance Mining (Marlon Dumas)

11:05 - 11:15 Break

11:15 - 11:45 Challenges in Information Visualization (Jack van Wijk)

11:45 - 12:15 An Overview of the Research Collaboration between TU/e and QUT in the area of Process Mining (Arthur ter Hofstede)

Abstract: Metric learning and model interpretability

Metric learning aims at an automated adaptation of the distance measure which is used for the comparison of data points within a machine learning model. It does not only greatly enhance the capability of popular distance-based classifiers such as k-NN, clustering, or prototype based methods, but it also facilitates model interpretability under the umbrella of so-called relevance learning: the latter refers to the identification of the most popular features and feature correlations for a given task at hand; notable application range from biomedical data analysis up to the industrial process models. Within the talk, I will give an overview about recent results on metric learning for prototype based models. After a presentation of the general principle of metric learning, its theoretical background and applications, I will focus on two recent research directions which are of particular interest for applications: How to guarantee valid model interpretability in the case of high data dimensionality? How to link relevance learning to data visualisation?

Presented by Barbara Hammer

CoSeLoG Meeting (Dutch)

During the CoSeLoG meeting several participants in the CoSeLoG proejct present their vision on collaboration withing processes and process standardisation.

The CoSeLoG meeting will we fully in Dutch and the schedule is as follows:

13:30-13:35 Opening

13:35-14:00 Processtandaardisatie in gemeenteland, droom of werkelijkheid? (Gemeenten Bergeijk, Eersel, en Oirschot)

14:00-14:25 Moeder en dochters: een gecompliceerde relatie. Werken met moedersjablonen in processtandaardisatie. (Dimpact)

14:25-14:40 Koffie & Thee

14:40-15:05 Een aanpak om de generatiekloof tussen moeders en dochters te dichten. (Dennis Schunselaar)

15:05-15:30 Cloudprocessen werken beter. (Perceptive Software)

Defense Joos

At 16:00 hours the defense of Joos will start in Auditorium 4 of Eindhoven University of Technology. After a brief presentation of 10 minutes of his work, the committee will ask Joos to defend his work and thesis. The whole defense is in English.

A digital version of the thesis can be downloaded here (PDF, 8 MB)

Summary Thesis

The goal of process mining is to automatically produce process models that accurately describe processes by considering only an organization’s records of its operational processes. Such records are typically captured in the form of event logs, consisting of cases and events related to these cases. Using these event logs process models can be discovered. Over the last decade, many such process discovery techniques have been developed, producing process models in various forms, such as Petri nets, BPMN-models, EPCs, YAWL-models etc. Furthermore, many authors have compared these techniques by focusing on the properties of the models produced, while at the same time the applicability of various techniques have been compared in case-studies. In this thesis we present a new process discovery algorithm: the Evolutionary Tree Miner, or ETM for short. The Evolutionary Tree Miner however has some unique characteristics and features, that are not found in existing process discovery algorithms. The main property of the Evolutionary Tree Miner is that it always produces a sound (i.e., syntactically correct) process model. Although this is a prerequisite for the process model to be used for further analysis, very few of the existing process discovery algorithms can guarantee this.

Another main feature of the Evolutionary Tree Miner is that it is a flexible algorithm. The four well known quality dimensions in process discovery (replay fitness, precision, generalization and simplicity) are explicitly incorporated in the Evolutionary Tree Miner. Additional quality metrics can be easily added to the Evolutionary Tree Miner. The Evolutionary Tree Miner is able to balance the different provided quality metrics and is able to produce process models that have a specific balance of these quality dimensions, as specified by the user.

The third main feature of the Evolutionary Tree Miner is that it is easily extensible. In this thesis we discuss several scenarios where the Evolutionary Tree Miner can be applied. We discuss the following extensions and applications in more detail in this thesis:

1. The discovery of a collection of process models, each having a unique and superior set of characteristics for the provided quality dimensions (more concretely: the process models are Pareto optimal).

2. Discovery of a process model given a (collection of) normative process models. This allows for the repair of a process model, using the observed behavior.

3. Discovery of a configurable process model that describes multiple event logs, for instance from different organizations, using only one process model and a configuration of the model for each event log.

4. A comparison framework is discussed that allows for the comparison of executions of similar processes. This framework is able to replay the behavior of one organization on the (configurable) process model of another, to provide insights into differences and commonalities between the process execution of the different organizations.

The Evolutionary Tree Miner is implemented as a plug-in for the process mining toolkit ProM. Furthermore, the usage of the different plug-ins created in this thesis is discussed. Additionally, it is shown how the Evolutionary Tree Miner can be extended to include custom quality dimensions.

The Evolutionary Tree Miner, and all of its extensions, are evaluated using both artificial and real-life data sets.

Reception

After the defense, the decision of the committee and the verdict follows a reception to thank all that attended.

Registration

All presentations and the defense itself are public, but registration beforehand is appreciated. Please use the following Google Form: https://docs.google.com/forms/d/1GPKp5-WWDLxwYjgjt0RQJTHP7CNsXcLc0xb95Y7uPbQ/viewform?usp=send_form

 
defensejoos.txt · Last modified: 2014/10/14 17:03 by jbuijs     Back to top
Recent changes RSS feed Creative Commons License Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki Design by Chirripó